CVPR 2015 Papers

Object recognition by zero-shot learning (ZSL) aims to recognise objects without seeing any visual examples by learning knowledge transfer between seen and unseen object classes. This is typically achieved by exploring a semantic embedding space such as attribute space or semantic word vector space. In such a space, both seen and unseen class labels, as well as image features can be embedded (projected), and the similarity between them can thus be measured directly. Existing works differ in what embedding space is used and how to project the visual data into the semantic embedding space. Yet, they all measure the similarity in the space using a conventional distance metric (e.g. cosine) that does not consider the rich intrinsic structure, i.e. semantic manifold, of the semantic categories in the embedding space. In this paper we propose to model the semantic manifold in an embedding space using a semantic class label graph. The semantic manifold structure is used to redefine the distance metric in the semantic embedding space for more effective ZSL. The proposed semantic manifold distance is computed using a novel absorbing Markov chain process (AMP), which has a very efficient closed-form solution. The proposed new model improves upon and seamlessly unifies various existing ZSL algorithms. Extensive experiments on both the large scale ImageNet dataset and the widely used Animal with Attribute (AwA) dataset show that our model outperforms significantly the state-of-the-arts.

This paper contributes to automatic classification and localization of human actions in video. Whereas motion is the key ingredient in modern approaches, we assess the benefits of having objects in the video representation. Rather than considering a handful of carefully selected and localized objects, we conduct an empirical study on the benefit of encoding 15,000 object categories for action using 6 datasets totaling more than 200 hours of video and covering 180 action classes. Our key contributions are i) the first in-depth study of encoding objects for actions, ii) we show that objects matter for actions, and are often semantically relevant as well. iii) We establish that actions have object preferences. Rather than using all objects, selection is advantageous for action recognition. iv) We reveal that object-action relations are generic, which allows to transferring these relationships from the one domain to the other. And, v) objects, when combined with motion, improve the state-of-the-art for both action classification and localization.

Scaling machine learning methods to very large datasets has attracted considerable attention in recent years, thanks to easy access to ubiquitous sensing and data from the web. We study face recognition and show that three distinct properties have surprising effects on the transferability of deep convolutional networks (CNN): (1) The bottleneck of the network serves as an important transfer learning regularizer, and (2) in contrast to the common wisdom, performance saturation may exist in CNN's (as the number of training samples grows); we propose a solution for alleviating this by replacing the naive random subsampling of the training set with a bootstrapping process. Moreover, (3) we find a link between the representation norm and the ability to discriminate in a target domain, which sheds lights on how such networks represent faces. Based on these discoveries, we are able to improve face recognition accuracy on the widely used LFW benchmark, both in the verification (1:1) and identification (1:N) protocols, and directly compare, for the first time, with the state of the art Commercially-Off-The-Shelf system and show a sizable leap in performance.

This paper addresses the problem of clustering a very large number of photos (i.e. hundreds of millions a day) in a stream into millions of clusters. This is particularly important as the popularity of photo sharing websites, such as Facebook, Google, and Instagram. Given large number of photos available online, how to efficiently organize them is an open problem. To address this problem, we propose to cluster the binary hash codes of a large number of photos into binary cluster centers. We present a fast binary k-means algorithm that works directly on the similarity-preserving hashes of images and clusters them into binary centers on which we can build hash indexes to speedup computation. The proposed method is capable of clustering millions of photos on a single machine in a few minutes. We show that this approach is usually several magnitude faster than standard k-means and produces comparable clustering accuracy. In addition, we propose an online clustering method based on binary k-means that is capable of clustering large photo stream on a single machine, and show applications to spam detection and trending photo discovery.

Image semantic segmentation is the task of partitioning image into several regions based on semantic concepts. In this paper, we learn a weakly supervised semantic segmentation model from social images whose labels are not pixel-level but image-level; furthermore, these labels might be noisy. We present a joint conditional random field model leveraging various contexts to address this issue. More specifically, we extract global and local features in multiple scales by convolutional neural network and topic model. Inter-label correlations are captured by visual contextual cues and label co-occurrence statistics. The label consistency between image-level and pixel-level is finally achieved by iterative refinement. Experimental results on two real-world image datasets PASCAL VOC2007 and SIFT-Flow demonstrate that the proposed approach outperforms state-of-the-art weakly supervised methods and even achieves accuracy comparable with fully supervised methods.

Weakly supervised object detection, is a challenging task, where the training procedure involves learning at the same time both, the model appearance and the object location in each image. The classical approach to solve this problem is to consider the location of the object of interest in each image as a latent variable and minimize the loss generated by such latent variable during learning. However, as learning appearance and localization are two interconnected tasks, the optimization is not convex and the procedure can easily get stuck in a poor local minimum, the algorithm "misses" the object in some images. In this paper, we help the optimization to get close to the global minimum by enforcing a "soft" similarity between each possible location in the image and a reduced set of "exemplars", or clusters, learned with a convex formulation in the training images. The help is effective because it comes from a different and smooth source of information that is not directly connected with the main task. Results show that our method improves a strong baseline based on convolutional neural network features by more than 4 points without any additional features or extra computation at testing time but only adding a small increment of the training time due to the convex clustering.

We consider the problem of localizing unseen objects in weakly labeled image collections. Given a set of images annotated at the image level, our goal is to localize the object in each image. The novelty of our proposed work is in addition to building object appearance model from the weakly labeled data, we also make use of existing detectors for some other object classes (which we call ``familiar objects''). We propose a method for transferring the appearance models of the familiar objects to the unseen object. Our experimental results on both image and video datasets demonstrate the effectiveness of our approach.

We focus on modeling human activities comprising multiple actions in a completely unsupervised setting. Our model learns the high-level action co-occurrence and temporal relations between the actions in the activity video. We consider the video as a sequence of short-term action clips, called action-words, and an activity is about a set of action-topics indicating which actions are present in the video. Then we propose a new probabilistic model relating the action-words and the action-topics. It allows us to model long-range action relations that commonly exist in the complex activity, which is challenging to capture in the previous works. We apply our model to unsupervised action segmentation and recognition, and also to a novel application that detects forgotten actions, which we call action patching. For evaluation, we also contribute a new challenging RGB-D activity video dataset recorded by the new Kinect v2, which contains several human daily activities as compositions of multiple actions interacted with different objects. The extensive experiments show the effectiveness of our model.

We present a semi-supervised approach that localizes multiple unknown object instances in long videos. We start with a handful of labeled boxes and iteratively learn and label hundreds of thousands of object instances. We propose criteria for reliable object detection and tracking for constraining the semi-supervised learning process and minimizing semantic drift. Our approach does not assume exhaustive labeling of each object instance in any single frame, or any explicit annotation of negative data. Working in such a generic setting allow us to tackle multiple object instances in video, many of which are static. In contrast, existing approaches either do not consider multiple object instances per video, or rely heavily on the motion of the objects present. The experiments demonstrate the effectiveness of our approach by evaluating the automatically labeled data on a variety of metrics like quality, coverage (recall), diversity, and relevance to training an object detector.

The estimation of material properties is important for scene understanding, with many applications in vision, robotics, and structural engineering. This paper connects fundamentals of vibration mechanics with computer vision techniques in order to infer material properties from small, often imperceptible motion in video. Objects tend to vibrate in a set of preferred modes. The shapes and frequencies of these modes depend on the structure and material properties of an object. Focusing on the case where geometry is known or fixed, we show how information about an object's modes of vibration can be extracted from video and used to make inferences about that object's material properties. We demonstrate our approach by estimating material properties for a variety of rods and fabrics by passively observing their motion in high-speed and regular framerate video.

Visual saliency is a fundamental problem in both cognitive and computational sciences, including computer vision. In this paper, we discover that a high-quality visual saliency model can be learned from multiscale features extracted using deep convolutional neural networks(CNN), which have had many successes in visual recognition tasks. For learning such saliency models, we introduce a neural network architecture, which has fully connected layers on top of CNNs responsible for extracting features at three different scales. Our learned saliency model is capable of achieving state-of-the-art performance on all public benchmarks. We then propose a refinement method to enhance the spatial coherence of our saliency results. Finally, we point out that aggregating multiple saliency maps computed for different levels of image segmentation can further boost the performance, yielding saliency maps better than those generated from a single region decomposition. To promote further research and evaluation of visual saliency models, we also construct a large database of 4447 challenging images and their pixelwise saliency annotation. Experimental results demonstrate that our proposed method significantly outperforms all existing saliency estimation techniques, improving the F-Measure by 5.0% and 13.2% respectively on the MSRA-B dataset and our new dataset, and lowering the mean absolute error by 5.7% and 35.1% respectively on the same two datasets.

In this work, we formulate a new weakly supervised domain generalization problem for the visual recognition task by using loosely labeled web images/videos as training data. Specifically, we aim to address two challenging issues when learning robust classifiers: 1) enhancing the generalization capability of the learnt classifiers to any unseen target domain; and 2) coping with noise in the labels of training web images/videos in the source domain. To address the first issue, we assume the training web images/videos may come from multiple hidden domains with different data distributions. We then extend the multi-class SVM formulation to learn one classifier for each class and each latent domain such that multiple classifiers from each class can be effectively integrated to achieve better generalization capability. To address the second issue, we partition the training samples in each class into multiple clusters. By treating each cluster as a "bag" and the samples in each cluster as "instances", we formulate a new multi-instance learning (MIL) problem for domain generalization by selecting a subset of training samples from each training bag and simultaneously learning the optimal classifiers based on the selected samples. Moreover, we also extend our newly proposed Weakly Supervised Domain Generalization (WSDG) approach by taking advantage of the additional textual descriptions that are only available in the training web images/videos as privileged information. Extensive experiments on four benchmark datasets demonstrate the effectiveness of our new approaches for visual recognition by learning from web data.

Many visual recognition problems can be approached by counting instances. To determine whether an event is present in a long internet video, one could count how many frames seem to contain the activity. Classifying the activity of a group of people can be done by counting the actions of individual people. Encoding these cardinality relationships can reduce sensitivity to clutter, in the form of irrelevant frames or individuals not involved in a group activity. Learned parameters can encode how many instances tend to occur in a class of interest. To this end, this paper develops a powerful and flexible framework to infer any cardinality relation between latent labels in a multi-instance model. Hard or soft cardinality relations can be encoded to tackle diverse levels of ambiguity. Experiments on tasks such as human activity recognition, video event detection, and video summarization demonstrate the effectiveness of using cardinality relations for improving recognition results.

How can we know whether a statement about our world is valid. For example, given a relationship between a pair of entities e.g., `eat(horse, hay)', how can we know whether this relationship is true or false in general. Gathering such knowledge about entities and their relationships is one of the fundamental challenges in knowledge extraction. Most previous works on knowledge extraction have focused purely on text-driven reasoning for verifying relation phrases. In this work, we introduce the problem of visual verification of relation phrases and developed a Visual Knowledge Extraction system called VisKE. Given a verb-based relation phrase between common nouns, our approach assess its validity by jointly analyzing over text and images and reasoning about the spatial consistency of the relative configurations of the entities and the relation involved. Our approach involves no explicit human supervision thereby enabling large-scale analysis. Using our approach, we have already verified over 12000 relation phrases. Our approach has been used to not only enrich existing textual knowledge bases by improving their recall, but also augment open-domain question-answer reasoning.

All that structure from motion algorithms "see" are sets of 2D points. We show that these impoverished views of the world can be faked for the purpose of reconstructing objects in challenging settings, such as from a single image, or from a few ones far apart, by recognizing the object and getting help from a collection of images of other objects from the same class. We synthesize virtual views by computing geodesics on novel networks connecting objects with similar viewpoints, and introduce techniques to increase the specificity and robustness of factorization-based object reconstruction in this setting. We report accurate object shape reconstruction from a single image on challenging PASCAL VOC data, which suggests that the current domain of applications of rigid structure-from-motion techniques may be significantly extended.

We characterize the problem of pose estimation for rigid objects in terms of determining viewpoint to explain coarse pose and keypoint prediction to capture the finer details. We address both these tasks in two different settings - the constrained setting with known bounding boxes and the more challenging detection setting where the aim is to simultaneously detect and correctly estimate pose of objects. We present Convolutional Neural Network based architectures for these and demonstrate that leveraging viewpoint estimates can substantially improve local appearance based keypoint predictions. In addition to achieving significant improvements over state-of-the-art in the above tasks, we analyze the error modes and effect of object characteristics on performance to guide future efforts towards this goal.

We present a novel method for summarizing raw, casually captured videos. The objective is to create a short summary that still conveys the story. It should thus be both, interesting and representative for the input video. Previous methods often used simplified assumptions and only optimized for one of these goals. Alternatively, they used hand-defined objectives that were optimized sequentially by making consecutive hard decisions. This limits their use to a particular setting. Instead, we introduce a new method that (i) uses a supervised approach in order to learn the importance of global characteristics of a summary and (ii) jointly optimizes for multiple objectives and thus creates summaries that posses multiple properties of a good summary. Experiments on two challenging and very diverse datasets demonstrate the effectiveness of our method, where we outperform or match current state-of-the-art.

Video magnification reveals subtle variations that would be otherwise invisible to the naked eye. Current techniques require all motion in the video to be very small, which is unfortunately not always the case. Tiny yet meaningful motions are often combined with larger motions, such as the small vibrations of a gate as it rotates, or the microsaccades in a moving eye. We present a layer-based video magnification approach that can amplify small motions within large ones. An examined region/layer is temporally aligned and subtle variations are magnified. Matting is used to magnify only region of interest while maintaining integrity of nearby sites. Results show handling larger motions, larger amplification factors and significant reduction in artifacts over state of the art.

Video event recognition still faces great challenges due to large intra-class variation and low image resolution, in particular for surveillance videos. To mitigate these challenges and to improve the event recognition performance, various context information from the feature level, the semantic level, as well as the prior level is utilized. Different from most existing context approaches that utilize context in one of the three levels through shallow models like support vector machines, or probabilistic models like BN and MRF, we propose a deep hierarchical context model that simultaneously learns and integrates context at all three levels, and holistically utilizes the integrated contexts for event recognition. We first introduce two types of context features describing the event neighborhood, and then utilize the proposed deep model to learn the middle level representations and combine the bottom feature level, middle semantic level and top prior level contexts together for event recognition. The experiments on state of art surveillance video event benchmarks including VIRAT 1.0 Ground Dataset, VIRAT 2.0 Ground Dataset, and the UT-Interaction Dataset demonstrate that the proposed model is quite effective in utilizing the context information for event recognition. It outperforms the existing context approaches that also utilize multiple level contexts on these event benchmarks.